Cross-scenario question-answering framework for non-performing assets based on knowledge graph

Abstract

The present invention provides a cross-scenario question-and-answer framework for non-performing assets based on a knowledge graph, which involves the fields of artificial intelligence and legal technology. The framework includes a knowledge fusion layer that sets a multi-source data access module for acquiring heterogeneous data from legal and regulatory databases, judicial case libraries, corporate financial reporting systems, and asset trading platforms; configures a legal entity parsing engine that converts unstructured legal texts into entity-relationship-attribute triples that can be represented by a knowledge graph based on ontology mapping technology; sets a financial data standardization component that converts financial data in different formats into a unified semantic representation through predefined financial indicator mapping rules; the knowledge fusion layer of this framework can integrate multi-source heterogeneous data such as laws and regulations, financial data, and business processes through a multi-source data access module, a legal entity parsing engine, and a financial data standardization component.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and legal technology, and specifically to a cross-scenario question-answering framework for non-performing assets based on knowledge graphs. Background Art

[0002] With the continuous advancement of artificial intelligence and big data technologies, the demand for intelligent and efficient solutions in the NPL disposal sector is becoming increasingly urgent. Accurately understanding complex business scenarios and rapidly conducting risk assessments and making decisions are key to improving disposal efficiency and reducing risks.

[0003] According to Chinese patent application number 202410113798.4, a legal question-and-answer system based on intent recognition and knowledge graphs is disclosed. It includes a data storage module, a data processing module, a dialogue management module, a natural language understanding module, a model API interface service module, a knowledge computation module, and a front-end interaction module. The data storage module manages Elasticsearch and Neo4j databases; the data processing module collects data through web crawlers and transmits it to the data storage module after processing; the dialogue management module configures the question-and-answer system's operating mode and distributes question processing tasks; the natural language understanding module analyzes the intent of question statements; the model API interface service module connects the dialogue module with the natural language understanding module; the knowledge computation module selects the appropriate question retrieval method based on the configuration and generates query statements for retrieval; and the front-end interaction module receives user question input and displays the answer feedback. This invention provides a convenient way for the public to obtain legal advice.

[0004] Although the above technologies have made some progress in legal question answering and text understanding, the following problems still exist in the actual application of cross-scenario question answering and disposal of non-performing assets:

[0005] Problem 1: Inadequate knowledge integration and management. Existing systems often focus on single-domain knowledge processing, making it difficult to integrate heterogeneous data from multiple sources, including laws and regulations, financial data, and business processes. Knowledge representation often relies on simple matching or shallow analysis, making it difficult to accurately understand the relationships between complex legal concepts and multi-source knowledge. Furthermore, imperfect knowledge update mechanisms prevent timely adaptation to regulatory changes and the dynamic demands of business scenarios, making it difficult to build a comprehensive and accurate knowledge system.

[0006] The second problem is limited reasoning and decision-making capabilities. Traditional reasoning methods lack support for multi-step reasoning and explainability, making it difficult to conduct in-depth analysis and make effective decisions when faced with the complex legal, financial, and market scenarios involved in the disposal of non-performing assets. Furthermore, insufficient reasoning efficiency and adaptability make it difficult to quickly respond to the diverse needs of different scenarios, making it difficult to meet the timeliness and accuracy requirements for risk assessment and disposal plan development in actual business operations.

[0007] Problem three: Inadequate system application performance. The existing system's architectural design lacks scalability, making it difficult to cope with the ever-increasing data volume and business complexity in the non-performing asset sector. Its low concurrency performance makes it unable to meet the needs of multiple users, such as financial institutions and asset management companies, to use it simultaneously. Its incomplete security mechanisms make it difficult to ensure the security of sensitive information related to non-performing assets during data transmission, storage, and processing, making it difficult to meet the practical requirements of large-scale, high-reliability applications.

[0008] Therefore, a cross-scenario question-answering framework for non-performing assets based on knowledge graph is needed to solve the above problems. Summary of the Invention

[0009] Technical problems solved

[0010] In response to the shortcomings of the existing technology, the present invention provides a cross-scenario question-answering framework for non-performing assets based on knowledge graphs, which solves the problems in the following background technologies.

[0011] Technical Solution

[0012] To achieve the above objectives, the present invention is implemented through the following technical solutions: a knowledge graph-based non-performing asset cross-scenario question-answering framework, including a framework that includes a knowledge fusion layer, a dynamic reasoning layer, a scenario adaptation layer, a decision verification layer, a knowledge evolution module, a cross-scenario migration module, and a federated knowledge learning architecture. Each layer works together through a standardized service interface. The specific composition and data flow method are as follows:

[0013] The knowledge fusion layer is used to integrate heterogeneous data and construct a unified knowledge graph, including a multi-source data access module, a legal entity parsing engine, a financial data standardization component, a business process knowledge module and a knowledge fusion engine, wherein: the multi-source data access module is configured to obtain heterogeneous data from a legal and regulatory database, a judicial case library, an enterprise financial reporting system and an asset trading platform; the legal entity parsing engine converts unstructured legal texts into entity-relationship-attribute triples that can be represented by a knowledge graph based on ontology mapping technology; the financial data standardization component converts financial data in different formats into a unified semantic representation through predefined financial indicator mapping rules; the business process knowledge module analyzes historical disposal cases based on process mining technology and extracts standardized non-performing asset disposal process templates; the knowledge fusion engine uses graph embedding technology to achieve semantic alignment of legal, financial and business knowledge subgraphs to generate a fused knowledge graph;

[0014] The dynamic reasoning layer is used to process user queries and generate preliminary answers, including a query understanding module, a multi-strategy reasoning engine, a three-dimensional risk assessment model and an answer generation module, wherein: the query understanding module converts user natural language queries into semantic query expressions executable by the knowledge graph; the multi-strategy reasoning engine performs cross-entity multi-hop reasoning, supporting rule reasoning, case reasoning and graph reasoning; the three-dimensional risk assessment model calculates the legal risk index, market volatility index and asset liquidity index based on the entity relationship network in the knowledge graph; and the answer generation module generates a preliminary answer containing a chain of evidence.

[0015] Preferably, the scenario adaptation layer is used to adjust the answer output according to the user context, and includes a scenario recognition module, a cross-scenario knowledge transfer module, an answer reconstruction engine, and a multimodal display module, wherein: the scenario recognition module automatically matches the corresponding handling scenario template based on the user context information; the cross-scenario knowledge transfer module identifies reusable knowledge components between different scenarios; the answer reconstruction engine converts the preliminary answer into a structured output according to the scenario template requirements; the multimodal display module supports three output forms: text, chart, and knowledge graph visualization;

[0016] The decision verification layer is used to verify the accuracy and feasibility of the answer, including a multi-dimensional evaluation module, a disposal solution executable verification module, an evidence chain tracing engine, and a feedback learning module. Among them, the multi-dimensional evaluation module verifies the legal compliance, financial feasibility, and process integrity of the answer based on predefined evaluation indicators; the disposal solution executable verification module predicts the resource consumption and time cost of the disposal process through simulation technology; the evidence chain tracing engine records the complete reasoning path from query to answer; and the feedback learning module updates the relationship weights and reasoning rules of the knowledge graph based on user feedback.

[0017] The knowledge evolution module is used to dynamically update and optimize the knowledge graph, including a knowledge update detection unit, a knowledge incremental update engine, a knowledge conflict resolution module, and a knowledge graph dynamic evolution mechanism, wherein: the knowledge update detection unit monitors the update of the data source in real time; the knowledge incremental update engine integrates the newly added knowledge into the existing knowledge graph; the knowledge conflict resolution module resolves conflicts in the knowledge graph;

[0018] The cross-scenario transfer module is used to support the transfer of knowledge between different scenarios, and includes a scenario feature extraction unit, a meta-learning engine, a solution verification unit, and a cross-modal knowledge representation module. The scenario feature extraction unit extracts features from scenario-related information; the meta-learning engine learns common knowledge representations between different scenarios; and the solution verification unit verifies the effectiveness of the cross-scenario transfer solution.

[0019] The federated knowledge learning architecture is used for distributed knowledge processing and includes local knowledge processing nodes, a security aggregation center, a privacy protection mechanism, and an institutional credit assessment module. The local knowledge processing nodes preprocess and extract features from local data; the security aggregation center aggregates knowledge uploaded by each node; and the privacy protection mechanism ensures privacy during data transmission and aggregation.

[0020] Each functional module of the framework is implemented through a standardized microservice interface, supporting independent deployment and horizontal expansion. The data flow between each layer adopts a message queue mechanism, and the message queue has different topics to distinguish different types of data, including knowledge fusion data topics, reasoning request data topics, and answer output data topics. The system as a whole supports high-concurrency query processing.

[0021] The legal entity parsing engine of the knowledge fusion layer includes a legal concept extraction unit, a relationship extraction unit, and an attribute labeling unit. The legal concept extraction unit uses a pre-trained deep learning model, fine-tuned with legal domain corpus, to identify legal concepts in legal texts. The relationship extraction unit extracts relationships between legal concepts based on a neural network combined with a legal ontology library. The attribute labeling unit completes the labeling of legal entity attributes by combining rule templates with semi-supervised learning.

[0022] The financial data standardization component includes a variety of financial indicator mapping rules, covering assets, liabilities, equity, income and expense accounts, and supports data format parsing, conversion and verification;

[0023] The disposal process templates extracted by the business process knowledge module include asset inventory, value assessment, judicial auction, debt restructuring and bankruptcy liquidation. Each template contains process steps, participating entities, execution standards and time nodes.

[0024] Preferably, the multi-strategy reasoning engine of the dynamic reasoning layer includes a rule reasoning unit, a case reasoning unit, a graph reasoning unit and a reasoning fusion unit, wherein: the rule reasoning unit is based on the legal, financial and business rule base and adopts forward and reverse reasoning; the case reasoning unit matches historical disposal cases through a similarity algorithm; the graph reasoning unit uses a graph neural network to model the relationship of the knowledge graph; and the reasoning fusion unit integrates the reasoning results through a weighted mechanism;

[0025] The three-dimensional risk assessment model calculates the legal risk index, market volatility index and asset liquidity index by weighted calculation, and adjusts the weights based on the closeness and frequency of entity relationships; the chain of evidence includes at least five reasoning steps, including query parsing, subgraph retrieval, rule matching, case comparison and result integration.

[0026] Preferably, the scene recognition module of the scene adaptation layer adopts a multimodal fusion model to calculate the matching probability of each scene template based on the text information input by the user, historical query records, current time and geographical location;

[0027] The scenario templates of the scenario adaptation layer include a judicial auction scenario template, a debt restructuring scenario template, and a bankruptcy liquidation scenario template. Each template contains multiple structured sub-templates, and each sub-template defines at least 10 standardized fields. The sub-templates of the judicial auction scenario template include an auction asset information sub-template, an auction process stage sub-template, a participant information sub-template, an auction result record sub-template, and a risk warning sub-template. The auction asset information sub-template contains fields for asset name, asset category, assessed value, starting price, and title certificate number.

[0028] The multimodal display module uses different graphic layouts and color coding to distinguish entity and relationship types when visualizing the knowledge graph.

[0029] Preferably, the multi-dimensional evaluation module of the decision verification layer includes a legal compliance evaluation unit, a financial feasibility evaluation unit, and a process integrity evaluation unit, wherein: the legal compliance evaluation unit matches legal clauses with disposal plans through a rule engine and a legal knowledge graph to determine compliance; the financial feasibility evaluation unit evaluates the financial risks of disposal plans based on simulation methods; and the process integrity evaluation unit verifies the integrity of the disposal process through process modeling technology;

[0030] The disposal scheme executability verification module is based on discrete event simulation technology to construct a simulation model including resource, activity and event elements to predict resource consumption and time cost.

[0031] Preferably, the knowledge update detection unit of the knowledge evolution module adopts an incremental detection algorithm to monitor the updates of data sources such as the legal and regulatory database and the enterprise financial reporting system in real time;

[0032] The incremental knowledge update engine uses graph embedding technology to integrate new knowledge into the existing knowledge graph;

[0033] The knowledge conflict resolution module resolves conflicts through priority sorting and negotiation strategies based on the conflict resolution strategy library;

[0034] The dynamic evolution mechanism of the knowledge graph is based on a reinforcement learning algorithm, which adjusts the structure and parameters of the knowledge graph according to the treatment effect.

[0035] Preferably, the scene feature extraction unit of the cross-scene migration module uses a neural network to extract features from scene text information and knowledge graph structure;

[0036] The meta-learning engine learns common knowledge representations across different scenarios based on a model-independent meta-learning algorithm.

[0037] The solution verification unit verifies the effectiveness of the cross-scenario migration solution by comparing simulation with actual cases;

[0038] The cross-modal knowledge representation module converts unstructured judicial documents and evaluation reports into semantic vectors that can be represented by a knowledge graph.

[0039] Preferably, the local knowledge processing node of the federated knowledge learning architecture performs preprocessing and feature extraction on the local non-performing asset data;

[0040] The security aggregation center aggregates the knowledge uploaded by each node through a secure multi-party computing protocol;

[0041] The privacy protection mechanism uses encryption technology to ensure privacy security during data transmission and aggregation;

[0042] The institutional credit assessment module is based on blockchain technology and records the knowledge contribution and data quality of each participating institution.

[0043] Preferably, the framework supports cross-jurisdictional reasoning and includes a multi-jurisdictional knowledge subgraph, a legal conflict identification engine, a conflict resolution decision module, and a legal knowledge transfer module. The legal conflict identification engine compares legal clauses in knowledge subgraphs across different jurisdictions to identify potential conflicts. The conflict resolution decision module, based on legal analogy reasoning technology, transfers the handling experience of one jurisdiction to another.

[0044] The legal knowledge transfer module is based on legal analogy reasoning technology and supports cross-jurisdictional knowledge transfer.

[0045] Preferably, the framework is configured with a system support module. When the framework uses the system support module, each functional module of the framework is interconnected through a standardized microservice interface, supporting independent deployment and horizontal expansion. The data flow between each layer adopts the Kafka message queue mechanism, and the message queue is set with a knowledge fusion data topic, a reasoning request data topic, and an answer output data topic. The framework is equipped with an intelligent prompt module.

[0046] The intelligent prompt module adopts collaborative filtering algorithm and association rule mining technology to recommend legal terms and disposal strategies based on the user's historical query behavior and current disposal scenario; the system as a whole supports the processing of 1,000 concurrent user query requests.

[0047] Beneficial effects

[0048] This invention provides a cross-scenario question-answering framework for non-performing assets based on knowledge graphs. It has the following beneficial effects:

[0049] 1. The framework's knowledge fusion layer integrates heterogeneous data from multiple sources, including laws, regulations, financial data, and business processes, through a multi-source data access module, a legal entity resolution engine, and financial data standardization components. Leveraging technologies such as ontology mapping and graph embedding, the framework achieves deep knowledge fusion and semantic alignment, addressing the challenges of insufficient knowledge fusion and management in existing technologies and constructing a comprehensive, accurate, and dynamically updated knowledge graph. The knowledge evolution module, based on timestamp detection and reinforcement learning mechanisms, updates the knowledge graph in real time, adapting promptly to regulatory changes and the dynamic demands of business scenarios, providing a solid data foundation for the disposal of non-performing assets.

[0050] 2. The dynamic reasoning layer of this invention is equipped with a multi-strategy reasoning engine and a three-dimensional risk assessment model. It uses a combination of rule-based reasoning, case-based reasoning, and graph-based reasoning to achieve cross-entity multi-hop reasoning and in-depth analysis. It also provides interpretability support through evidence chain generation, effectively addressing the limitations of traditional reasoning methods in reasoning and decision-making in the disposal of non-performing assets. In complex legal, financial, and market scenarios, it can quickly and accurately conduct risk assessments and formulate disposal plans, improving the timeliness and accuracy of decision-making to meet actual business needs.

[0051] 3. The scenario adaptation layer of the present invention is based on multimodal fusion scenario recognition technology and meta-learning algorithms, which can automatically match the disposal scenario templates, realize the rapid reconstruction and multimodal display of answers, and significantly improve the system's adaptability to different scenarios. At the same time, the framework adopts a federated knowledge learning architecture and standardized microservice interface design, combines Docker containers and Kubernetes to achieve independent deployment and horizontal expansion, and cooperates with the Kafka message queue mechanism to greatly improve the system's concurrent processing capabilities and data flow efficiency. In addition, the security aggregation center and privacy protection mechanism use homomorphic encryption, blockchain and other technologies to ensure the security of data during transmission and storage, comprehensively solve the problem of insufficient application performance of existing systems, and can meet the large-scale, high-reliability application needs of non-performing asset disposal. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a question-and-answer flow chart of the present invention;

[0053] Figure 2 This is a diagram of the question-answering framework of the present invention;

[0054] Figure 3 This is an evaluation simulation diagram of the present invention. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Specific embodiment one:

[0057] like Figure 1-3 As shown in the figure, a cross-scenario question-answering framework for non-performing assets based on knowledge graphs includes a knowledge fusion layer, a dynamic reasoning layer, a scenario adaptation layer, and a decision verification layer. Each layer works together through a standardized service interface. The specific composition and data flow are as follows:

[0058] The knowledge fusion layer is used to integrate heterogeneous data and build a unified knowledge graph, including a multi-source data access module, a legal entity parsing engine, a financial data standardization component, a business process knowledge module and a knowledge fusion engine, wherein: the multi-source data access module is configured to obtain heterogeneous data from legal and regulatory databases, judicial case libraries, corporate financial reporting systems and asset trading platforms; the legal entity parsing engine converts unstructured legal texts into entity-relationship-attribute triples that can be represented by knowledge graphs based on ontology mapping technology; the financial data standardization component converts financial data in different formats into a unified semantic representation through predefined financial indicator mapping rules; the business process knowledge module analyzes historical disposal cases based on process mining technology and extracts standardized non-performing asset disposal process templates; the knowledge fusion engine uses graph embedding technology to implement method Semantic alignment of legal, financial, and business knowledge subgraphs to generate a fused knowledge graph. The knowledge fusion layer sets up a multi-source data access module to obtain heterogeneous data from legal and regulatory databases, judicial case libraries, corporate financial reporting systems, and asset trading platforms. A legal entity parsing engine is configured to convert unstructured legal texts into entity-relationship-attribute triples that can be represented by the knowledge graph based on ontology mapping technology. A financial data standardization component is set up to convert financial data in different formats into a unified semantic representation through predefined financial indicator mapping rules. A business process knowledge module is constructed to conduct process mining on historical disposal cases based on Petri net theory and extract standardized disposal process templates. Through the knowledge fusion engine, graph embedding technology is used to achieve semantic alignment of legal, financial, and business ternary knowledge subgraphs to generate a fused knowledge graph.

[0059] The dynamic reasoning layer is used to process user queries and generate preliminary answers, including a query understanding module, a multi-strategy reasoning engine, a three-dimensional risk assessment model and an answer generation module, among which: the query understanding module converts user natural language queries into semantic query expressions that can be executed by the knowledge graph; the multi-strategy reasoning engine performs cross-entity multi-hop reasoning, supporting rule reasoning, case reasoning and graph reasoning; the three-dimensional risk assessment model calculates the legal risk index, market volatility index and asset liquidity index based on the entity relationship network in the knowledge graph; the answer generation module generates preliminary answers containing a chain of evidence. The dynamic reasoning layer sets a query understanding module to convert user natural language queries into semantic query expressions that can be executed by the knowledge graph; configures a multi-strategy reasoning engine to perform cross-entity multi-hop reasoning; constructs a three-dimensional risk assessment model, calculates the legal risk index, market volatility index and asset liquidity index based on the entity relationship network in the knowledge graph; sets an answer generation module to generate preliminary answers containing a chain of evidence;

[0060] The scenario adaptation layer is used to adjust the answer output according to the user context, including a scenario recognition module, a cross-scenario knowledge transfer module, an answer reconstruction engine and a multimodal display module, among which: the scenario recognition module automatically matches the corresponding handling scenario template based on user context information; the cross-scenario knowledge transfer module identifies reusable knowledge components between different scenarios; the answer reconstruction engine converts the preliminary answer into a structured output according to the scenario template requirements; the multimodal display module supports three output forms: text, chart and knowledge graph visualization. The scenario adaptation layer sets the scenario recognition module to automatically match the corresponding handling scenario template based on user context information; configures the answer reconstruction engine to convert the preliminary answer into a structured output according to the formatting requirements of the scenario template; and constructs a multimodal display module to support three output forms: text, chart and knowledge graph visualization.

[0061] The decision verification layer is used to verify the accuracy and feasibility of the answer, including a multi-dimensional evaluation module, a disposal plan executable verification module, an evidence chain tracing engine and a feedback learning module, among which: the multi-dimensional evaluation module verifies the legal compliance, financial feasibility and process integrity of the answer based on predefined evaluation indicators; the disposal plan executable verification module predicts the resource consumption and time cost of the disposal process through simulation technology; the evidence chain tracing engine records the complete reasoning path from query to answer; the feedback learning module updates the relationship weights and reasoning rules of the knowledge graph based on user feedback. Decision verification layer: set up a multi-dimensional evaluation module, based on a predefined evaluation indicator system, to verify the legal compliance, financial feasibility and process integrity of the preliminary answer; configure the evidence chain tracing engine to record the complete reasoning path from query to answer; build a feedback learning module to update the relationship weights and reasoning rules in the knowledge graph based on user feedback data.

[0062] In the legal entity parsing engine of the knowledge fusion layer, the legal concept extraction unit adopts a model that combines bidirectional encoder representation with conditional random fields. It identifies legal concepts in legal texts by pre-training and fine-tuning on massive legal text corpora; the relationship extraction unit adopts a bidirectional long short-term memory network based on the attention mechanism, combined with the legal field ontology library, to extract the relationship between legal concepts; the attribute labeling unit uses a combination of rule-based template matching and semi-supervised learning to complete the legal entity attribute labeling. In the financial data standardization component of the knowledge fusion layer, 128 financial indicator mapping rules cover financial subjects such as assets, liabilities, owner's equity, income, and expenses. Each rule includes source data format analysis, target data format analysis, and target data format analysis. The specific algorithms for format conversion and data verification realize the standardized conversion of data from different corporate financial reporting systems through regular expression matching and semantic mapping functions. The standardized disposal process templates extracted by the business process knowledge module of the knowledge fusion layer include asset inventory, value assessment, judicial auction, debt restructuring, and bankruptcy liquidation. Each template contains elements such as process steps, participating entities, execution standards, and time nodes. Among them, the asset inventory template includes steps such as asset inventory and ownership verification; the value assessment template includes content such as evaluation method selection and value calculation; the judicial auction template includes links such as auction announcement release and bidding process; the debt restructuring template includes processes such as program design and negotiation; and the bankruptcy liquidation template includes stages such as asset liquidation and debt repayment.

[0063] In the multi-strategy reasoning engine of the dynamic reasoning layer, the rule reasoning unit employs a hybrid reasoning strategy combining forward and backward reasoning to reason based on a pre-set legal, financial, and business rule base. The case reasoning unit utilizes a case retrieval algorithm based on cosine similarity to match similar cases from the historical case library. The graph reasoning unit employs a graph attention network model to model and reason about complex relationships within the knowledge graph. The reasoning fusion unit integrates the results of the three reasoning units through a weighted voting mechanism. In the three-dimensional risk assessment model of the dynamic reasoning layer, the legal risk index is calculated by assigning different weights to different types of legal violations and taking a weighted sum based on factors such as the number of violations and the amount involved. The market volatility index is calculated based on an autoregressive integrated moving average model of time series analysis, after predicting and analyzing asset transaction price data. The asset liquidity index is calculated by constructing an assessment model that includes indicators such as asset turnover rate and days to liquidity. The risk level is divided into five levels based on the weighted total score of the three indices. The evidence chain at the dynamic reasoning layer consists of at least five consecutive reasoning steps and their knowledge sources. These steps include parsing user query intent, retrieving knowledge graph subgraphs, matching reasoning rules, comparing relevant cases, and integrating results. The knowledge sources are specifically data records from legal and regulatory databases, judicial case libraries, corporate financial reporting systems, and asset trading platforms. Furthermore, when calculating each index, the three-dimensional risk assessment model adjusts the calculation weights based on factors such as the closeness and frequency of relationships between entities in the knowledge graph. During the evidence chain generation process, each reasoning step is evaluated for credibility, and the evaluation results serve as a reference for answer generation.

[0064] The scenario templates of the scenario adaptation layer include judicial auction scenario template, debt restructuring scenario template and bankruptcy liquidation scenario template; each scenario template contains 5 structured sub-templates, and each sub-template defines at least 10 standardized fields. The structured sub-template of the judicial auction scenario template includes auction asset information sub-template, auction process stage sub-template, participant information sub-template, auction result record sub-template, and risk warning sub-template. The auction asset information sub-template contains fields such as asset name, asset category, assessed value, starting price, and title certificate number. The scene recognition module of the scene adaptation layer automatically matches the corresponding handling scene template based on user context information, adopts a model based on multimodal fusion, integrates context information such as user input text information, historical query records, current time, geographic location, etc., and matches by calculating the matching probability of each scene template. The cross-scene knowledge transfer module of the scene adaptation layer identifies reusable knowledge components between different scene templates based on the meta-learning algorithm. Specifically, it constructs a meta-learning model, trains on historical data of multiple scenes, extracts common features and knowledge patterns between scenes, and realizes the migration of knowledge components. In the scene adaptation layer, the answer reconstruction engine filters, classifies and formats the data according to the characteristics of different scene templates when converting preliminary answers into structured outputs; when visualizing the knowledge graph, the multimodal display module uses different graphic layouts and color coding methods to distinguish different types of entities and relationships.

[0065] The multi-dimensional evaluation modules of the decision verification layer include a legal compliance evaluation unit, a financial feasibility evaluation unit, and a process integrity evaluation unit. The legal compliance evaluation unit adopts a combination of a rule engine and a legal knowledge graph to judge legal compliance by matching legal clauses with legal elements in the disposal plan; the financial feasibility evaluation unit conducts risk assessment on the financial indicators in the disposal plan based on the Monte Carlo simulation method; the process integrity evaluation unit models and analyzes the disposal process through the Petri net model to verify the integrity of the process; the disposal plan executable verification module of the decision verification layer simulates the disposal process execution process based on discrete event simulation technology, and predicts resource consumption and time cost by constructing a simulation model containing resources, activities, events and other elements. The decision verification layer configures an evidence chain tracing engine to record the complete reasoning path from query to answer; and constructs a feedback learning module to update the relationship weights and reasoning rules in the knowledge graph based on user feedback data.

[0066] The framework also includes a knowledge evolution module, which is equipped with a knowledge update detection unit, a knowledge incremental update engine and a knowledge conflict resolution module; the knowledge update detection unit adopts a timestamp-based incremental detection algorithm to monitor the updates of data sources such as the legal and regulatory database and the corporate financial reporting system in real time; the knowledge incremental update engine adopts a knowledge fusion algorithm based on graph embedding to integrate the new knowledge with the existing knowledge graph; the knowledge conflict resolution module is based on a conflict resolution strategy library and adopts priority sorting and negotiation strategies to resolve knowledge conflicts; the framework's knowledge graph dynamic evolution mechanism automatically adjusts the structure and parameters of the knowledge graph according to the disposal effect based on the reinforcement learning algorithm, and the knowledge evolution module is used to dynamically update and optimize the knowledge graph, including a knowledge update detection unit, a knowledge incremental update engine, a knowledge conflict resolution module and a knowledge graph dynamic evolution mechanism, among which: the knowledge update detection unit monitors the updates of the data source in real time; the knowledge incremental update engine integrates the new knowledge into the existing knowledge graph; the knowledge conflict resolution module resolves conflicts in the knowledge graph.

[0067] The framework also includes a cross-scenario migration module, which is equipped with a scene feature extraction unit, a meta-learning engine and a solution verification unit; the scene feature extraction unit uses a convolutional neural network to extract features from scene text information, and combines the graph convolutional network to extract features from the scene-related knowledge graph structure information; the meta-learning engine learns the common knowledge representation between different scenarios based on the model-independent meta-learning algorithm; the solution verification unit verifies the effectiveness of the cross-scenario migration solution by comparing simulation with actual cases; the cross-modal knowledge representation module converts unstructured judicial documents and evaluation reports into semantic vectors that can be represented by the knowledge graph. The cross-scenario migration module is used to support the transfer of knowledge between different scenarios, including a scene feature extraction unit, a meta-learning engine, a solution verification unit and a cross-modal knowledge representation module, wherein: the scene feature extraction unit extracts features from scene-related information; the meta-learning engine learns the common knowledge representation between different scenarios; the solution verification unit verifies the effectiveness of the cross-scenario migration solution.

[0068] The federated knowledge learning architecture is used for distributed knowledge processing, including local knowledge processing nodes, a security aggregation center, a privacy protection mechanism and an institutional credit assessment module, among which: the local knowledge processing nodes preprocess and extract features of local data; the security aggregation center aggregates the knowledge uploaded by each node; the privacy protection mechanism ensures privacy security during data transmission and aggregation. The framework adopts a federated knowledge learning architecture, sets up local knowledge processing nodes, a security aggregation center and a privacy protection mechanism; the local knowledge processing nodes preprocess and extract features of local non-performing asset data; the security aggregation center aggregates the knowledge uploaded by each local knowledge processing node through a secure multi-party computing protocol; the privacy protection mechanism adopts homomorphic encryption technology to ensure the privacy security of data during transmission and aggregation; the institutional credit assessment module records the knowledge contribution and data quality of each participating institution based on blockchain technology.

[0069] The framework supports cross-jurisdictional reasoning, and sets up multi-jurisdictional knowledge subgraphs, a legal conflict identification engine, and a conflict resolution decision module; the legal conflict identification engine identifies potential legal conflicts by comparing the legal clauses and rules in the knowledge subgraphs of different jurisdictions; the conflict resolution decision module transfers the handling experience of one jurisdiction to another to resolve legal conflicts based on legal analogy reasoning technology; the legal knowledge transfer module transfers the handling experience of one jurisdiction to another based on legal analogy reasoning technology.

[0070] The framework configures system support modules so that each functional module can be interconnected through standardized microservice interfaces, supporting independent deployment and horizontal expansion; the data flow between layers adopts the Kafka message queue mechanism, and the message queue sets different topics to distinguish different types of data, including knowledge fusion data topics, reasoning request data topics, and answer output data topics; each functional module implements standardized microservices through the RESTful API interface, and each microservice can be independently deployed through a Docker container and horizontally expanded through Kubernetes; the system as a whole supports query request processing of 1,000 concurrent users. The intelligent prompt module uses collaborative filtering algorithms and association rule mining technology based on the user's historical query behavior and current disposal scenarios to actively recommend possible legal terms and disposal strategies.

[0071] The knowledge graph-based cross-scenario question-answering framework for non-performing assets achieves efficient answers to non-performing asset-related questions and scientific evaluation of disposal plans through the collaborative operation of the knowledge fusion layer, dynamic reasoning layer, scenario adaptation layer, and decision verification layer. Each layer is closely linked and interconnected. The specific working method is as follows:

[0072] Knowledge Fusion Layer: The multi-source data access module serves as the data entry point. Leveraging interface protocols tailored to the characteristics of different data sources, such as database connection protocols and APIs, it regularly or in real time collects heterogeneous data from legal and regulatory databases, judicial case repositories, corporate financial reporting systems, and asset trading platforms. This data covers a wide range of content, including legal texts, judicial decisions, corporate balance sheets, and asset transaction records. This raw data is then transmitted to subsequent modules within the knowledge fusion layer for in-depth processing.

[0073] The legal entity parsing engine analyzes unstructured legal texts, identifying legal concepts using a model combining bidirectional encoder representation and conditional random fields. A bidirectional long-short-term memory network based on an attention mechanism extracts relationships between concepts. Attribute labeling is accomplished through a combination of rule-based template matching and semi-supervised learning, ultimately transforming legal texts into entity-relationship-attribute triples. The financial data standardization component uses 128 predefined mapping rules covering multiple financial categories to unify financial data in various formats into a standard semantic representation through regular expression matching, semantic mapping function conversion, and data validation algorithms. The business process knowledge module mines historical disposal cases based on Petri net theory to extract standardized disposal process templates for asset inventory and valuation, clarifying the process steps, participating entities, and other elements of each template.

[0074] The knowledge fusion engine uses graph embedding technology to semantically align the legal, financial, and business knowledge subgraphs, eliminating semantic differences and generating a fused knowledge graph. This fused knowledge graph is not only the final product of the knowledge fusion layer but also the core data foundation for subsequent reasoning and analysis in the dynamic reasoning layer, providing a rich and structured knowledge base for the operation of the entire framework.

[0075] Dynamic Reasoning Layer: After receiving natural language queries from users, the query understanding module uses natural language processing technology to convert them into executable semantic query expressions within the knowledge graph. This expression is then passed to the multi-strategy reasoning engine. The rule reasoning unit within the multi-strategy reasoning engine employs a combination of forward and backward reasoning strategies based on a pre-defined legal, financial, and business rule base. The case reasoning unit uses a cosine similarity-based case retrieval algorithm to match similar cases from the historical case library. The graph reasoning unit employs a graph attention network model to model and reason about complex relationships within the knowledge graph. The reasoning fusion unit integrates the results of these three components through a weighted voting mechanism.

[0076] The three-dimensional risk assessment model is based on the knowledge graph entity relationship network and combines the entity association information in the fused knowledge graph generated by the knowledge fusion layer to calculate the legal risk index, market volatility index, and asset liquidity index. Risk levels are then classified based on the weighted total score. The weights are dynamically adjusted during the calculation process based on the closeness and frequency of the associations between entities. Based on the inference results and risk assessment information, the answer generation module generates a preliminary answer consisting of at least five consecutive inference steps and their knowledge sources. The knowledge source of each inference step points to the data source processed by the knowledge fusion layer. The inference steps are also evaluated for credibility and incorporated into the answer. Once the preliminary answer is formed, it is transmitted to the scenario adaptation layer for further processing.

[0077] Scene adaptation layer:

[0078] The scene recognition module adopts a model based on multimodal fusion, integrating contextual information such as user input text information and historical query records, calculates the matching probability of each scene template (judicial auction, debt restructuring, bankruptcy liquidation, etc.), and automatically selects the most matching disposal scene template from the standardized disposal process template and other knowledge provided by the knowledge fusion layer.

[0079] The answer reconstruction engine filters, categorizes, and formats the preliminary answers generated by the dynamic reasoning layer according to the formatting requirements of the selected scenario template, converting them into structured output. The multimodal presentation module supports three output formats: text, charts, and knowledge graph visualization, presenting the processed answers to users in intuitive and diverse ways. The cross-scenario knowledge transfer module, based on a meta-learning algorithm, trains a meta-learning model on historical data from multiple scenarios to extract common features and knowledge patterns and identify reusable knowledge components. When encountering new scenarios, relevant knowledge can be transferred from the knowledge fusion layer's knowledge system, improving the framework's adaptability. The cross-scenario knowledge transfer module in the scenario adaptation layer uses a meta-learning algorithm to identify reusable knowledge components across different scenario templates. Specifically, by building a meta-learning model and training it on historical data from multiple scenarios, it extracts common features and knowledge patterns across scenarios and enables the transfer of knowledge components. When encountering new scenarios, relevant knowledge can be transferred from the knowledge fusion layer's knowledge system, improving the framework's adaptability. Processed answers are passed to the decision verification layer for verification, if necessary.

[0080] Decision Verification Layer: The multi-dimensional assessment module includes a legal compliance assessment unit, a financial feasibility assessment unit, and a process integrity assessment unit. These modules verify the answers or solutions output by the scenario adaptation layer from legal, financial, and process perspectives, respectively, based on a predefined evaluation index system. The legal compliance assessment unit uses a rule engine combined with a legal knowledge graph to match legal clauses with the legal elements of the solution to determine compliance. The financial feasibility assessment unit uses Monte Carlo simulation to assess the financial indicator risks of the solution. The process integrity assessment unit verifies the integrity of the solution process using a Petri net model. The solution executable verification module in the decision verification layer simulates the solution execution process using discrete event simulation technology. By constructing a simulation model that includes resources, activities, and events, it predicts resource consumption and time costs. This module can proactively identify potential problems encountered in the actual implementation of the solution, providing a basis for solution optimization and ensuring the high executable nature of the final solution. The evidence chain traceability engine records the complete reasoning path from query to answer. A feedback learning module is built to update the relationship weights and reasoning rules in the knowledge graph based on user feedback data.

[0081] Other modules: The knowledge evolution module works closely with the knowledge fusion layer. The knowledge update detection unit adopts a timestamp-based incremental detection algorithm to monitor data source updates in real time. Once an update is found, the knowledge incremental update engine adopts a graph embedding-based knowledge fusion algorithm to integrate the new knowledge into the existing knowledge graph. The knowledge conflict resolution module resolves knowledge conflicts based on the conflict resolution strategy library. The knowledge graph dynamic evolution mechanism adjusts the graph structure and parameters according to the disposal effect based on the reinforcement learning algorithm, providing a more accurate knowledge basis for the dynamic reasoning layer.

[0082] The cross-scenario migration module's scenario feature extraction unit uses convolutional neural networks and graph convolutional networks to extract scene text and knowledge graph structural features. The meta-learning engine learns general knowledge representations based on a model-independent meta-learning algorithm. The solution verification unit verifies the effectiveness of the migration solution through simulations and comparisons with real-world cases. This process involves accessing and processing knowledge from the knowledge fusion layer. The cross-modal knowledge representation module converts unstructured judicial documents and assessment reports into semantic vectors representable in the knowledge graph, enriching the knowledge content of the knowledge fusion layer.

[0083] The framework adopts a federated knowledge learning architecture. Local knowledge processing nodes pre-process and extract features of local non-performing asset data. The security aggregation center aggregates the knowledge uploaded by each node through a secure multi-party computing protocol. The privacy protection mechanism uses homomorphic encryption technology to ensure data security. The institutional credit assessment module uses blockchain technology to record the knowledge contribution and data quality of each participating institution. This architecture provides a safe and reliable environment for data collection and processing in the knowledge fusion layer.

[0084] The framework supports cross-jurisdictional reasoning. The legal conflict identification engine compares the legal clauses and rules in the multi-jurisdictional knowledge subgraph to identify potential legal conflicts. The conflict resolution decision module transfers the handling experience of one jurisdiction to another to resolve conflicts based on legal analogy reasoning technology, relying on the multi-jurisdictional knowledge subgraph constructed by the knowledge fusion layer.

[0085] The framework's functional modules are implemented through standardized microservice interfaces, based on RESTful APIs, independently deployed using Docker containers, and horizontally scalable through Kubernetes. Kafka message queues are used for data flow between layers, with distinct themes for distinguishing data types, such as knowledge fusion data, inference request data, and answer output data. The system as a whole supports processing 1,000 concurrent user query requests. The intelligent prompt module, based on historical user query behavior and current action scenarios, utilizes collaborative filtering algorithms and association rule mining techniques to extract relevant legal provisions and action strategies from the knowledge system of the knowledge fusion layer and proactively recommend them. Specific embodiment two:

[0087] like Figure 1-3 As shown, the key algorithms mentioned in Example 1 are analyzed in detail below, including their core mathematical formulas and explanations:

[0088] Graph embedding technology for the knowledge fusion layer:

[0089] Graph embedding technology is used to map entities and relationships in knowledge graphs into low-dimensional vector spaces for semantic alignment and fusion. One of the commonly used methods is the TransE model, whose core formula is as follows:

[0090]

[0091] in: are the embedding vectors of the head entity, relation, and tail entity respectively; represents the L2 norm; is a scoring function used to measure the triple The rationality of . The loss function is usually defined as:

[0092]

[0093] in: is the correct set of triples, is a set of negative sample triplets; is the interval hyperparameter.

[0094] Explanation and problem solving:

[0095] Formula Function: TransE Assumption Relationship is a de novo entity To the end entity The translation operation, that is By optimizing the scoring function, entities and relationships are embedded into a unified vector space, maintaining semantic consistency. Problem Solved: In the knowledge fusion layer, the semantics of legal, financial, and business knowledge subgraphs vary significantly (for example, "debt restructuring" in legal text may have a different meaning than "debt restructuring" in financial statements). Using graph embedding technology, the framework aligns heterogeneous knowledge subgraphs into a unified semantic space, generating a fused knowledge graph.

[0096] Multi-strategy reasoning in the dynamic reasoning layer (graph reasoning unit):

[0097] The graph reasoning unit uses the graph attention network (GAT) to perform complex relationship modeling and reasoning. Its core formula is as follows:

[0098]

[0099]

[0100] in: is a node and The initial embedding vector of is a learnable weight matrix; is the weight vector of the attention mechanism; Represents vector concatenation; is a node For neighbor nodes The attention weight of is a node The set of neighbor nodes of is the activation function (such as ReLU); is a node Updated embedding vector.

[0101] Explanation and problem solving:

[0102] Formula function: GAT dynamically calculates the contribution weight of each neighbor node to the current node through the attention mechanism , and update the node representation according to the weighted aggregation This allows the framework to capture complex multi-hop relationships in knowledge graphs.

[0103] Problem Solved: In the disposal of non-performing assets, the relationships between entities are complex (for example, Company A's debt restructuring may involve legal risks, financial data, and market fluctuations). GAT uses multi-hop reasoning to uncover deep relationships (e.g., "Company A - Debt Restructuring - Legal Risk - Involved - Certain Regulations").

[0104] Three-dimensional risk assessment model of the dynamic reasoning layer:

[0105] Calculate the legal risk index using a three-dimensional risk assessment model , Market Volatility Index , asset liquidity index , and weighted fusion into the total risk level. The core formula is as follows:

[0106] 1. Legal risk index:

[0107]

[0108] in: It is the weight of quasi-legal violations; is the number of occurrences of the violation; It is the amount involved.

[0109] 2. Market Volatility Index (based on ARIMA model prediction):

[0110]

[0111] in: is the actual transaction price; is the transaction price predicted by the ARIMA model; is the time window length; is the variance of the prediction error.

[0112] 3. Asset liquidity index:

[0113]

[0114] Where: TR is the asset turnover rate; DT is the days to cash; is the weight.

[0115] 4. Overall risk level:

[0116]

[0117] in: is the weighting coefficient, satisfying .

[0118] Explanation and problem solving:

[0119] Formula function: Legal risk index Quantify legal risks by comprehensively considering the frequency and amount of violations.

[0120] Market Volatility Index Use the ARIMA model to predict asset price fluctuations and quantify market risks.

[0121] Asset Liquidity Index Comprehensively evaluate asset turnover rate and days to cash to assess asset liquidity.

[0122] Overall risk level Weighted fusion of three-dimensional index to divide risk levels.

[0123] Problem Solved: Disposal of non-performing assets requires a comprehensive risk assessment involving multiple factors, including legal, market, and liquidity factors. The formula provides a quantitative risk assessment method to support scientific decision-making.

[0124] Multimodal fusion model of the scene adaptation layer:

[0125] The scene recognition module uses a multimodal fusion model to integrate text, history, time, and location information to calculate the scene template matching probability. Its core formula is as follows:

[0126]

[0127]

[0128] in: , , , is the feature vector of text, history, time and location; It is a vector concatenation operation; , , , are learnable weights and biases; It is The matching probability of a scene template.

[0129] Explanation and problem solving:

[0130] Formula function: The multimodal fusion model combines the features of different modalities (text, historical records, etc.) through a fully connected neural network and The function calculates the matching probability of each scene template.

[0131] Problem Solved: Disposal scenarios for non-performing assets vary, including judicial auctions and debt restructuring. Scenario templates need to be automatically matched based on user context. This formula enables the fusion of multimodal information and scenario matching.

[0132] Specific applications:

[0133] Extracting user input text features , Historical query record features , time characteristics , location features .

[0134] Concatenated feature vectors , calculate the matching probability through neural network .

[0135] Select the scenario template with the highest probability (such as the judicial auction template) for subsequent answer reconstruction.

[0136] Meta-learning of the scene adaptation layer (cross-scene knowledge transfer):

[0137] The cross-scenario knowledge transfer module is based on the meta-learning algorithm (MAML), and its core formula is as follows:

[0138]

[0139]

[0140] in: are the initial parameters of the model; are task-specific parameters; is the loss function of the task; is the learning rate; Is a collection of tasks.

[0141] Explanation and problem solving:

[0142] Function: MAML learns universal knowledge representation across scenarios through a two-stage optimization process: inner optimization and outer optimization. The inner optimization fine-tunes parameters for each scenario task, while the outer optimization updates global parameters, enabling the model to quickly adapt to new scenarios.

[0143] Problem Solved: Knowledge reuse across different scenarios (such as judicial auctions and debt restructuring) is difficult, and the framework needs to quickly adapt to new scenarios. Meta-learning extracts common features across scenarios to enable knowledge transfer. Specific embodiment three:

[0145] like Figure 1-3 As shown in the figure, the specific application logic steps of each module and algorithm in the knowledge graph-based non-performing asset cross-scenario question-answering framework are as follows:

[0146] Knowledge fusion layer:

[0147] The multi-source data access module uses database connection protocols and API interfaces to periodically or in real time acquire heterogeneous data from legal and regulatory databases, judicial case repositories, corporate financial reporting systems, and asset trading platforms. This data, including legal texts, judicial decisions, financial statements, and asset transaction records, is then temporarily stored pending processing. Within the legal entity resolution engine, the legal concept extraction unit uses a model combining bidirectional encoder representations with conditional random fields to identify legal concepts within legal texts based on pre-training and fine-tuning on a massive corpus of legal text. The relationship extraction unit utilizes a bidirectional long-short-term memory network based on an attention mechanism, combined with a legal domain ontology, to extract relationships between legal concepts. The attribute labeling unit uses a combination of rule-based template matching and semi-supervised learning to label legal entity attributes, forming entity-relationship-attribute triples. The financial data standardization component uses 128 predefined mapping rules covering financial categories such as assets and liabilities, matches source data formats through regular expressions, and applies semantic mapping functions and data validation algorithms to convert data from different corporate financial reporting systems into a unified semantic representation. The business process knowledge module conducts process mining on historical disposal cases based on Petri net theory, extracting standardized disposal process templates such as asset inventory and value assessment, and clarifying the process steps, participants, execution standards, and timelines for each template. The knowledge fusion engine utilizes graph embedding technology to semantically align the legal, financial, and business knowledge subgraphs, eliminating semantic differences and generating a fused knowledge graph to provide data support for subsequent reasoning.

[0148] Dynamic Inference Layer:

[0149] The query understanding module receives user natural language queries and, through natural language processing techniques such as word segmentation, part-of-speech tagging, and semantic analysis, converts them into semantic query expressions executable by the knowledge graph and passes them to the multi-strategy reasoning engine. Within the multi-strategy reasoning engine, the rule reasoning unit employs a combination of forward and reverse reasoning strategies based on a pre-set legal, financial, and business rule base. The case reasoning unit uses a case retrieval algorithm based on cosine similarity to match similar cases from the historical case library. The graph reasoning unit employs a graph attention network model to model and reason about complex relationships in the knowledge graph. The reasoning fusion unit integrates the results of the three reasoning units through a weighted voting mechanism. The three-dimensional risk assessment model is based on the knowledge graph entity relationship network and calculates the legal risk index, market volatility index, and asset liquidity index. The legal risk index is a weighted sum of the weights of different types of legal violations, the number of violations, and the amount. The market volatility index is derived by predicting and analyzing asset transaction price data using the autoregressive integrated moving average model of time series analysis. The asset liquidity index is calculated by constructing an assessment model that includes indicators such as asset turnover rate and days to liquidity. Five risk levels are divided according to the weighted total score of the three indices, and the weights are adjusted during the calculation based on the closeness and frequency of the association between entities. Based on the reasoning results and risk assessment information, the answer generation module generates a preliminary answer that includes at least five consecutive reasoning steps, such as user query intent analysis and knowledge graph subgraph retrieval, and their knowledge sources. The knowledge source points to a specific data source. At the same time, the credibility of each reasoning step is evaluated and incorporated into the answer. The preliminary answer is then transmitted to the scenario adaptation layer.

[0150] Scene adaptation layer:

[0151] The scenario recognition module uses a model based on multimodal fusion, integrating contextual information such as user input text, historical query records, current time, and geographic location. It calculates the matching probability of scenario templates such as judicial auctions, debt restructuring, and bankruptcy liquidation, and selects the template with the highest matching probability. The answer reconstruction engine filters, classifies, and formats preliminary answers according to the formatting requirements of the selected scenario template, converting them into structured output. The multimodal presentation module presents the structured output in three forms: text, charts, and knowledge graph visualization. Text directly displays content, charts are used to display numerical data, and knowledge graph visualization uses different graphical layouts and color coding to distinguish entities and relationships. The cross-scenario knowledge transfer module, based on a meta-learning algorithm, constructs a meta-learning model trained on historical data from multiple scenarios, extracts common features and knowledge patterns, identifies reusable knowledge components, and applies them in new scenarios. The processed answers are passed to the decision verification layer as needed.

[0152] Decision verification layer:

[0153] In the multi-dimensional assessment module, the legal compliance assessment unit uses a combination of a rule engine and a legal knowledge graph to match legal clauses with the legal elements of the disposal plan to determine compliance; the financial feasibility assessment unit uses the Monte Carlo simulation method to conduct a risk assessment of the financial indicators of the disposal plan; and the process integrity assessment unit uses the Petri net model to model and analyze the disposal process to verify its integrity. The disposal plan executable verification module uses discrete event simulation technology to construct a simulation model containing elements such as resources, activities, and events, simulate the execution process of the disposal process, and predict resource consumption and time costs. The evidence chain traceability engine records the complete reasoning path from query to answer, including data processing and reasoning processes at all levels. The feedback learning module collects user feedback data, updates the relationship weights and reasoning rules in the knowledge graph based on the feedback, optimizes the framework performance, and affects the subsequent knowledge fusion and reasoning process.

[0154] Other modules:

[0155] In the knowledge evolution module, the knowledge update detection unit uses a timestamp-based incremental detection algorithm to monitor data source updates in real time. The knowledge incremental update engine employs a graph embedding-based knowledge fusion algorithm to integrate new knowledge into the existing knowledge graph. The knowledge conflict resolution module, based on a conflict resolution strategy library, employs a priority ranking and negotiation strategy to resolve knowledge conflicts. The knowledge graph dynamic evolution mechanism, based on a reinforcement learning algorithm, adjusts the graph structure and parameters based on the effectiveness of the resolution. In the cross-scenario transfer module, the scenario feature extraction unit uses a convolutional neural network and a graph convolutional network to extract scenario text and knowledge graph structural features, respectively. The meta-learning engine learns general knowledge representations based on a model-independent meta-learning algorithm. The solution verification unit verifies the effectiveness of the cross-scenario transfer solution through simulation and real-world case comparisons. The cross-modal knowledge representation module converts unstructured judicial documents and assessment reports into semantic vectors representable in the knowledge graph. The framework adopts a federated knowledge learning architecture. Local knowledge processing nodes preprocess and extract features from local data. The security aggregation center aggregates knowledge through a secure multi-party computation protocol. The privacy protection mechanism uses homomorphic encryption to ensure data security. The institutional credit assessment module uses blockchain technology to record institutional knowledge contribution and data quality. The framework supports cross-jurisdictional reasoning. Its legal conflict identification engine compares legal clauses and rules across multiple legal domain knowledge subgraphs to identify conflicts. The conflict resolution decision module leverages legal analogical reasoning techniques to transfer experience and resolve conflicts. The framework's functional modules are implemented through standardized microservice interfaces, based on RESTful APIs and deployed using Docker containers and Kubernetes for horizontal scalability. Kafka message queues are used between layers, with different data topics configured for data flow. The system supports 1,000 concurrent user queries. The intelligent notification module utilizes collaborative filtering algorithms and association rule mining techniques based on historical user query behavior and current scenarios to extract and proactively recommend legal clauses and resolution strategies from the knowledge fusion layer.

[0156] The framework's system support module ensures the stable operation of the intelligent prompt module by providing a highly available and scalable infrastructure. The system adopts a standardized microservice architecture and interconnects the functional modules based on the RESTful API interface. Each microservice is independently deployed through a Docker container, and Kubernetes is used to achieve automated horizontal scaling and load balancing. Data flow relies on the Kafka message queue mechanism, setting up knowledge fusion data topics, reasoning request data topics, answer output data topics, and intelligent recommendation data topics to ensure real-time data processing and timely delivery of recommendation results. In addition, it is equipped with a distributed cache to accelerate access to users' historical query behaviors, and integrates Elasticsearch to support efficient association rule mining and recommendation algorithm calculations, jointly supporting the efficient query processing needs of 1,000 concurrent users, thereby providing reliable computing and storage support for the intelligent prompt module. Specific embodiment four:

[0158] like Figure 1-3 As shown, the following is a detailed hardware composition and hardware description of each module in Example 1:

[0159] The knowledge graph-based cross-scenario question-and-answer framework for non-performing assets utilizes a distributed cluster architecture, with the overall hardware environment consisting of server clusters, storage device clusters, and network equipment, within which each module plays a distinct role. The multi-source data access module in the knowledge fusion layer utilizes data acquisition servers equipped with high-performance multi-core CPUs (such as the Intel Xeon Gold series), 128GB or more of memory, and multiple network interfaces with speeds exceeding 10Gbps, paired with solid-state storage arrays (SSD arrays) as data cache devices. This enables efficient collection and temporary storage of multi-source heterogeneous data. Modules such as the legal entity resolution engine rely on a cluster of compute servers equipped with NVIDIA A100 GPUs, 256GB or more of memory, and NVMe SSDs, combined with the Ceph distributed storage system, to complete complex data processing and knowledge graph construction tasks.

[0160] The query understanding module of the dynamic reasoning layer uses a natural language processing server equipped with an AMD EPYC series multi-core CPU and more than 64GB of memory, and uses Intel Optane DC Persistent Memory as a local cache to quickly resolve user queries. Modules such as the multi-strategy reasoning engine rely on an inference computing cluster composed of servers equipped with NVIDIA H100 GPUs, which are interconnected through the InfiniBand high-speed network and equipped with an all-flash array to store data, ensuring efficient operation of inference computing.

[0161] Modules such as scene recognition in the scene adaptation layer use application servers equipped with Intel Core i9 series multi-core CPUs, more than 32GB of memory and SSD storage, and are paired with NVIDIA RTX 30 series GPUs to complete multimodal display tasks; the cross-scene knowledge transfer module uses training servers equipped with NVIDIA A40 GPUs and more than 128GB of memory, combined with the GlusterFS distributed file system, to achieve model training and data storage.

[0162] Modules such as the multi-dimensional evaluation of the decision verification layer rely on evaluation computing servers equipped with Intel Xeon Platinum series multi-core CPUs and more than 256GB of memory, and are paired with enterprise-level hard disk arrays (HDD arrays) to store data; modules such as the evidence chain tracing engine run the database management system through data management servers equipped with high-performance CPUs and more than 64GB of memory, and are backed up with tape libraries or cloud storage to ensure data management and storage security.

[0163] In other modules, the monitoring and update server of the knowledge evolution module is equipped with a multi-core CPU and more than 128GB of memory, and cooperates with a distributed version control system (such as Git distributed storage) to realize knowledge graph update management; the cross-scenario migration module shares hardware resources with related modules, and builds an independent test server cluster to verify the solution; the local knowledge processing nodes of the federated knowledge learning architecture are composed of edge servers or local workstations, and the security aggregation center adopts a high-performance server cluster, and deploys hardware encryption devices (such as encryption network cards, HSM) to ensure data security; the framework deployment and performance-related modules use x86 server clusters combined with virtualization technology, and use high-performance switches, routers and load balancing devices to build a network environment. A hybrid storage architecture (SSD and HDD combination) is adopted and data storage and management are realized through storage management software, which jointly support the efficient and stable operation of the framework. Specific embodiment five:

[0165] like Figure 1-3 As shown, the following provides specific use cases of the solution:

[0166] Case 1: Comprehensive Assessment of Judicial Auction of Enterprise Non-Performing Assets

[0167] A bank holds a company's non-performing assets and plans to dispose of them through judicial auction. Bank staff use this framework to conduct relevant analysis. At the knowledge fusion layer, the multi-source data access module retrieves the company's financial data, including assets, liabilities, and cash flow, from the company's financial reporting system. It also retrieves cases of similar corporate non-performing asset judicial auctions from the judicial case database, current market price data for similar assets from the asset trading platform, and legal provisions related to judicial auctions from the legal and regulatory database. After processing, this data is generated into a fused knowledge graph. At the dynamic reasoning layer, staff input "What are the risks and expected returns of the judicial auction of this company's non-performing assets?" The query understanding module converts this query into a semantic query expression. The multi-strategy reasoning engine combines rules, cases, and graph reasoning. The three-dimensional risk assessment model calculates the legal risk index (high due to the company's outstanding litigation), the market volatility index (medium due to the current economic situation), and the asset liquidity index (high due to the ease of auctioning assets). The answer generation module then generates a preliminary answer, including the reasoning steps and knowledge sources, indicating the risk of legal disputes associated with the auction and the uncertainty of expected returns due to market volatility. The scenario adaptation layer automatically matches judicial auction scenario templates. The answer reconstruction engine converts preliminary answers into structured output. The multimodal display module presents risk indicators in charts and textual descriptions of expected returns. The decision verification layer evaluates the answers. The legal compliance assessment unit verifies whether the auction process complies with legal regulations. The financial feasibility assessment unit analyzes the rationality of expected returns. The process integrity assessment unit verifies the completeness of the auction process steps. Ultimately, the answers are determined to be reliable, providing a basis for banks to formulate auction strategies.

[0168] Case 2: Formulation of a Non-Performing Asset Debt Restructuring Plan

[0169] An asset management company took over a portfolio of nonperforming assets and is considering debt restructuring. During the knowledge integration phase, the system integrates data on the financial distress of the companies involved in the nonperforming assets, industry trends, past successful and failed debt restructuring cases, and relevant laws and policies. When the asset manager asks, "How can I design a feasible debt restructuring plan for a particular company's nonperforming assets?" the dynamic reasoning layer responds quickly. The multi-strategy reasoning engine integrates knowledge from various sources, references debt restructuring cases of similar companies, and incorporates the current economic situation and legal regulations to infer multiple possible restructuring approaches. A three-dimensional risk assessment model assesses the risks of each approach and weighs the pros and cons. The scenario adaptation layer matches debt restructuring scenario templates and structures the inference results, presenting the key elements, expected outcomes, and potential risks of different restructuring plans in clear text and charts. The decision verification layer rigorously verifies the plans from the perspectives of legal compliance, financial feasibility, and process integrity, simulating potential issues that may arise during the restructuring process to ensure their feasibility. Ultimately, based on the system's proposed plan and the actual situation, the asset manager negotiates with the debtor company to finalize and implement a debt restructuring plan, successfully revitalizing the nonperforming assets.

[0170] Case 3: Cross-regional non-performing asset disposal strategy consulting

[0171] A large financial group holds nonperforming assets in multiple regions. Disposal is challenging due to differences in local laws, policies, and market environments. Using this framework, group employees input questions such as "What issues should be considered and what strategies should be adopted when disposing of similar nonperforming assets in different regions?" The knowledge fusion layer aggregates relevant data from each region to construct a multi-jurisdictional knowledge subgraph. The legal conflict identification engine in the dynamic reasoning layer compares the knowledge subgraphs across different jurisdictions, identifying differences in legal provisions and potential conflicts. The conflict resolution decision module, based on legal analogy reasoning, transfers successful disposal experiences from other regions. The multi-strategy reasoning engine, taking into account local circumstances, recommends disposal strategies for each region. For example, in Region A, where laws provide strong creditor protection, legal litigation may be recommended; in Region B, where the market is active, asset transfer may be prioritized. The scenario adaptation layer personalizes the recommended strategies based on the characteristics of each region. The decision verification layer verifies the feasibility and compliance of the strategies. Based on these recommendations, the group develops targeted cross-regional NPA disposal strategies to improve efficiency and success rates. Specific embodiment six:

[0173] like Figure 1-3 As shown, the complete experimental data is provided below:

[0174] Experimental objective: To verify the performance of the framework in cross-scenario question answering for non-performing assets, including knowledge graph construction efficiency, reasoning accuracy, risk assessment effect, scenario matching accuracy, and decision verification results.

[0175] Experimental scenarios: Three typical non-performing asset disposal scenarios are selected: judicial auction, debt restructuring, and bankruptcy liquidation.

[0176] Datasets: Legal and Regulatory Database: Contains 100,000 legal provisions. Judicial Case Database: Contains 50,000 historical disposal cases. Corporate Financial Reporting System: Contains 10,000 financial statements. Asset Trading Platform: Contains 20,000 transaction records.

[0177] Evaluation indicators: knowledge graph construction time (seconds), entity-relationship extraction accuracy (%), reasoning accuracy (%), reasoning time (seconds), risk assessment error (%), scene matching accuracy (%),

[0178] Decision verification pass rate (%), feedback learning optimization effect (% improvement in inference accuracy before and after weight update).

[0179] Experimental conditions:

[0180] Hardware: 16-core CPU, 64GB memory, GPU support.

[0181] Number of concurrent users: 1000 (in line with the technical solution's requirement to support 1000 concurrent users).

[0182] The table is as follows:

[0183] Table 1: Performance of knowledge fusion layer and dynamic reasoning layer:

[0184] Scenario Knowledge graph construction time (seconds) Entity-relationship extraction accuracy (%) Inference accuracy (%) Risk assessment error (%) Inference time (seconds) Judicial Auction 120 92.5 88.7 5.4 2.5 Debt restructuring 105 90.8 87.3 5.5 2.8 bankruptcy liquidation 135 93.2 89.1 5.2 2.3

[0185] Table 2: Performance of the scenario adaptation layer and decision verification layer:

[0186] Scenario Scene matching accuracy (%) Success rate of cross-scenario knowledge transfer (%) Verification pass rate (%) Feedback learning optimization effect (inference accuracy improvement %) Executability verification resource error (%) Judicial Auction 94.5 85.2 90.6 3.5 6.2 Debt restructuring 93.8 83.9 89.9 3.2 6.5 bankruptcy liquidation 95.1 86.7 91.3 3.8 5.9

[0187] In the above table:

[0188] The knowledge graph construction time is between 105 and 135 seconds, and the entity-relationship extraction accuracy is stable at above 90%, indicating that the framework can efficiently process heterogeneous data.

[0189] The inference accuracy is between 87% and 89%, the risk assessment error is controlled at 5.2% to 5.5%, and the inference time is 2.3 to 2.8 seconds, meeting real-time requirements.

[0190] The scene matching accuracy exceeds 93%, and the cross-scene knowledge transfer success rate is 83%-86%, showing that the framework has strong adaptability to multiple scenarios.

[0191] The verification pass rate is between 89% and 91%, the feedback learning optimization effect is about 3%, and the resource prediction error is between 5.9% and 6.5%, indicating that the decision support is reliable and optimizable.

[0192] Figure 3 This is a simulation chart for evaluating this method. The horizontal axis represents three scenarios: judicial auction, debt restructuring, and bankruptcy liquidation. The vertical axis represents the risk value (in units of risk index, ranging from 0 to 180). The legend includes the legal risk index (blue), market volatility index (orange), asset liquidity index (yellow), and total risk (purple).

[0193] From the figure we can conclude that:

[0194] The legal risk index of debt restructuring and bankruptcy liquidation is relatively high (about 160-170), significantly higher than that of judicial auction (about 40), indicating that legal risk is the main influencing factor.

[0195] Market volatility and asset liquidity indices are low (approximately 5-15), contributing less to the overall risk.

[0196] The total risk (purple) is lower in judicial auctions (about 70) than in debt restructuring and bankruptcy liquidation (about 80), reflecting the lower risk in the former.

[0197] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further restrictions, an element defined by the statement "comprising a reference structure" does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0198] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A cross-scenario question-answering framework for non-performing assets based on a knowledge graph, including a framework characterized by: The framework includes a knowledge fusion layer, a dynamic reasoning layer, a scenario adaptation layer, a decision verification layer, a knowledge evolution module, a cross-scenario migration module, and a federated knowledge learning architecture. Each layer works together through a standardized service interface. The specific composition and data flow are as follows: The knowledge fusion layer is used to integrate heterogeneous data and construct a unified knowledge graph, including a multi-source data access module, a legal entity parsing engine, a financial data standardization component, a business process knowledge module and a knowledge fusion engine, wherein: the multi-source data access module is configured to obtain heterogeneous data from a legal and regulatory database, a judicial case library, an enterprise financial reporting system and an asset trading platform; the legal entity parsing engine converts unstructured legal texts into entity-relationship-attribute triples that can be represented by a knowledge graph based on ontology mapping technology; the financial data standardization component converts financial data in different formats into a unified semantic representation through predefined financial indicator mapping rules; the business process knowledge module analyzes historical disposal cases based on process mining technology and extracts standardized non-performing asset disposal process templates; the knowledge fusion engine uses graph embedding technology to achieve semantic alignment of legal, financial and business knowledge subgraphs to generate a fused knowledge graph; The dynamic reasoning layer is used to process user queries and generate preliminary answers, including a query understanding module, a multi-strategy reasoning engine, a three-dimensional risk assessment model, and an answer generation module. Among them, the query understanding module converts user natural language queries into semantic query expressions that can be executed by the knowledge graph; the multi-strategy reasoning engine performs cross-entity multi-hop reasoning, supporting rule reasoning, case reasoning, and graph reasoning; the three-dimensional risk assessment model calculates the legal risk index, market volatility index, and asset liquidity index based on the entity relationship network in the knowledge graph; and the answer generation module generates a preliminary answer that includes a chain of evidence. The scenario adaptation layer is used to adjust the answer output according to the user context, and includes a scenario recognition module, a cross-scenario knowledge transfer module, an answer reconstruction engine, and a multimodal presentation module. The scenario recognition module automatically matches the corresponding handling scenario template based on user context information; the cross-scenario knowledge transfer module identifies reusable knowledge components between different scenarios; the answer reconstruction engine converts the preliminary answer into a structured output according to the scenario template requirements; and the multimodal presentation module supports three output formats: text, charts, and knowledge graph visualization. The decision verification layer is used to verify the accuracy and feasibility of the answer, including a multi-dimensional evaluation module, a disposal solution executable verification module, an evidence chain tracing engine, and a feedback learning module. Among them, the multi-dimensional evaluation module verifies the legal compliance, financial feasibility, and process integrity of the answer based on predefined evaluation indicators; the disposal solution executable verification module predicts the resource consumption and time cost of the disposal process through simulation technology; the evidence chain tracing engine records the complete reasoning path from query to answer; and the feedback learning module updates the relationship weights and reasoning rules of the knowledge graph based on user feedback. The knowledge evolution module is used to dynamically update and optimize the knowledge graph, including a knowledge update detection unit, a knowledge incremental update engine, a knowledge conflict resolution module, and a knowledge graph dynamic evolution mechanism, wherein: the knowledge update detection unit monitors the update of the data source in real time; the knowledge incremental update engine integrates the newly added knowledge into the existing knowledge graph; the knowledge conflict resolution module resolves conflicts in the knowledge graph; The cross-scenario transfer module is used to support the transfer of knowledge between different scenarios, and includes a scenario feature extraction unit, a meta-learning engine, a solution verification unit, and a cross-modal knowledge representation module. The scenario feature extraction unit extracts features from scenario-related information; the meta-learning engine learns common knowledge representations between different scenarios; and the solution verification unit verifies the effectiveness of the cross-scenario transfer solution. The federated knowledge learning architecture is used for distributed knowledge processing and includes local knowledge processing nodes, a security aggregation center, a privacy protection mechanism, and an institutional credit assessment module. The local knowledge processing nodes preprocess and extract features from local data; the security aggregation center aggregates knowledge uploaded by each node; and the privacy protection mechanism ensures privacy during data transmission and aggregation. Each functional module of the framework is implemented through a standardized microservice interface, supporting independent deployment and horizontal expansion. The data flow between each layer adopts a message queue mechanism, and the message queue has different topics to distinguish different types of data, including knowledge fusion data topics, reasoning request data topics, and answer output data topics. The system as a whole supports high-concurrency query processing. The legal entity parsing engine of the knowledge fusion layer includes a legal concept extraction unit, a relationship extraction unit, and an attribute labeling unit. The legal concept extraction unit uses a pre-trained deep learning model, fine-tuned with legal domain corpus, to identify legal concepts in legal texts. The relationship extraction unit extracts relationships between legal concepts based on a neural network combined with a legal ontology library. The attribute labeling unit completes the labeling of legal entity attributes by combining rule templates with semi-supervised learning. The financial data standardization component includes a variety of financial indicator mapping rules, covering assets, liabilities, equity, income and expense accounts, and supports data format parsing, conversion and verification; The disposal process templates extracted by the business process knowledge module include asset inventory, value assessment, judicial auction, debt restructuring and bankruptcy liquidation. Each template contains process steps, participating entities, execution standards and time nodes.

2. The knowledge graph-based cross-scenario question-answering framework for non-performing assets according to claim 1 is characterized by: The multi-strategy reasoning engine of the dynamic reasoning layer includes a rule reasoning unit, a case reasoning unit, a graph reasoning unit, and a reasoning fusion unit. The rule reasoning unit uses forward and reverse reasoning based on the legal, financial, and business rule bases; the case reasoning unit matches historical disposal cases through a similarity algorithm; the graph reasoning unit uses a graph neural network to model knowledge graph relationships; and the reasoning fusion unit integrates reasoning results through a weighted mechanism. The three-dimensional risk assessment model calculates the legal risk index, market volatility index and asset liquidity index by weighted calculation, and adjusts the weights based on the closeness and frequency of entity relationships; the chain of evidence includes at least five reasoning steps, including query parsing, subgraph retrieval, rule matching, case comparison and result integration.

3. The knowledge graph-based cross-scenario question-answering framework for non-performing assets according to claim 1 is characterized by: The scene recognition module of the scene adaptation layer adopts a multimodal fusion model to calculate the matching probability of each scene template based on the text information input by the user, historical query records, current time and geographical location; The scenario templates of the scenario adaptation layer include a judicial auction scenario template, a debt restructuring scenario template, and a bankruptcy liquidation scenario template. Each template contains multiple structured sub-templates, and each sub-template defines at least 10 standardized fields. The sub-templates of the judicial auction scenario template include an auction asset information sub-template, an auction process stage sub-template, a participant information sub-template, an auction result record sub-template, and a risk warning sub-template. The auction asset information sub-template contains fields for asset name, asset category, assessed value, starting price, and title certificate number. The multimodal display module uses different graphic layouts and color coding to distinguish entity and relationship types when visualizing the knowledge graph.

4. The knowledge graph-based cross-scenario question-answering framework for non-performing assets according to claim 1 is characterized by: The multi-dimensional evaluation module of the decision verification layer includes a legal compliance evaluation unit, a financial feasibility evaluation unit, and a process integrity evaluation unit. Among them, the legal compliance evaluation unit uses a rule engine and a legal knowledge graph to match legal clauses with disposal plans to determine compliance; the financial feasibility evaluation unit uses simulation methods to evaluate the financial risks of disposal plans; and the process integrity evaluation unit verifies the integrity of the disposal process through process modeling technology. The disposal scheme executability verification module is based on discrete event simulation technology to construct a simulation model including resource, activity and event elements to predict resource consumption and time cost.

5. The knowledge graph-based cross-scenario question-answering framework for non-performing assets according to claim 1 is characterized by: The knowledge update detection unit of the knowledge evolution module adopts an incremental detection algorithm to monitor the updates of data sources such as the legal and regulatory database and the enterprise financial reporting system in real time; The incremental knowledge update engine uses graph embedding technology to integrate new knowledge into the existing knowledge graph; The knowledge conflict resolution module resolves conflicts through priority sorting and negotiation strategies based on the conflict resolution strategy library; The dynamic evolution mechanism of the knowledge graph is based on a reinforcement learning algorithm, which adjusts the structure and parameters of the knowledge graph according to the treatment effect.

6. The knowledge graph-based cross-scenario question-answering framework for non-performing assets according to claim 1 is characterized by: The scene feature extraction unit of the cross-scene migration module uses a neural network to extract features from scene text information and knowledge graph structure; The meta-learning engine learns common knowledge representations across different scenarios based on a model-independent meta-learning algorithm. The solution verification unit verifies the effectiveness of the cross-scenario migration solution by comparing simulation with actual cases; The cross-modal knowledge representation module converts unstructured judicial documents and evaluation reports into semantic vectors that can be represented by a knowledge graph.

7. The knowledge graph-based cross-scenario question-answering framework for non-performing assets according to claim 1 is characterized by: The local knowledge processing node of the federated knowledge learning architecture performs preprocessing and feature extraction on local non-performing asset data; The security aggregation center aggregates the knowledge uploaded by each node through a secure multi-party computing protocol; The privacy protection mechanism uses encryption technology to ensure privacy security during data transmission and aggregation; The institutional credit assessment module is based on blockchain technology and records the knowledge contribution and data quality of each participating institution.

8. The knowledge graph-based cross-scenario question-answering framework for non-performing assets according to claim 1 is characterized by: The framework supports cross-jurisdictional reasoning and includes a multi-jurisdictional knowledge subgraph, a legal conflict identification engine, a conflict resolution decision module, and a legal knowledge transfer module. The legal conflict identification engine compares legal clauses in knowledge subgraphs across different jurisdictions to identify potential conflicts. The conflict resolution decision module uses legal analogy reasoning techniques to transfer experience from one jurisdiction to another. The legal knowledge transfer module is based on legal analogy reasoning technology and supports cross-jurisdictional knowledge transfer.

9. The knowledge graph-based cross-scenario question-answering framework for non-performing assets according to claim 1 is characterized by: The framework is equipped with a system support module. When using the system support module, each functional module of the framework is interconnected through a standardized microservice interface, supporting independent deployment and horizontal expansion. The data flow between each layer adopts the Kafka message queue mechanism, and the message queue is set with a knowledge fusion data topic, a reasoning request data topic, and an answer output data topic. The framework is equipped with an intelligent prompt module. The intelligent prompt module recommends legal terms and disposal strategies based on the user's historical query behavior and current disposal scenario using collaborative filtering algorithms and association rule mining technology.

Images